The application analyzes customer behavior and segments them by preference. The data based on collected information about purchased, frequency of purchases, date of last purchase, participation in promotions, etc. (a total of 30+ parameters). Then recommendations for each group are selected and the most favorable moments for sending messages are determined. Each client receives a mail with an offer customer is interested in, and at a time when he/she is most inclined to buy goods.
What Was Done
Using the collected data, we trained a neural network model. The model detects patterns in the data and remembers them. The quality of model training is evaluated by metrics. For example we make a prediction which of the buyers will come in a month and compare it with the actual buyers. The percentage of buyers for which the model made the right decision is usually more than 87%.
As a result of implementing this program, the client was able to manage the marketing budget effectively due to increase message conversion rate by 44% and reduce the cost of mailing by more than 2 times, and the investment in the development paid off in six months. In addition, the company was able to avoid annoying spam, increasing customer loyalty and getting a higher NPS index.
Technologies and tools: Python, Keras, CatBoost, Spark